Skip to main content

Edgify Architectures

Last Updated by Gil Ninveh on June 20th, 2024.

Overview​

In this section, we outline the 2 most common Edgify architectures. There are additional architectures which can be discussed if required (for example: the use of IP cameras / dealing with edge devices which fall below the minimum requirements).

Please consult your Sales Engineer, who will assist in customizing the best solution to suit your environment and needs.

Edgify Solution Components​

The following sections outline Edgify's components:

Device Agent​

The Agent is deployed on all devices. It enables devices to execute real-time predictions and continuously collect samples for training. SetupUI and Data Collector - Backend component within the agent which allows testing the AI model, collecting data and deleting images from the db.

Detector (Loss Prevention only)​

The Detector is deployed on all devices for Loss Prevention use-cases. It enables devices to execute real-time inference on Video feeds for analysis.

Trainer​

Worker​

The Worker is in charge of the Training process. In the Distributed architecture - the worker will be deployed on all edge devices and function as part of a collaborative training architecture, where all machines participate in the training process. In the Hybrid architecture - the worker will be installed on an Edgibox, which will store images and be the only machine to host the training process.

Collaborator​

The collaborative training module facilitates collaborative training between edge devices. It sits on edge devices.

Cross-site Collaborator​

Facilitates collaborative training between stores. Hosted in a datacenter or cloud.

Edgify Dashboard​

Edgify Dashboard enables Edgify to manage the Edgify system and provide AI collaborative training capabilities on edge devices. High-level capabilities:

  • Manage Devices
  • Data Management - Aggregative view of the data collected on all the edge devices
  • Run & Monitor training jobs
  • Save trained Models
  • Deploy and monitor Prediction Models

Edgibox​

Computer device used to host training components in environments with underpowered or incompatible edge devices. Can be provided by Edgify or produced by the customer provided it complies with the minimum requirements.